feat: allow to reset histograms

[vmarchaud]

I'm collecting runTime and waitTime metrics for my pool with otel:

const metrics = {
  workersQueueWaitTime: meter.createObservableGauge('content_workers_queue_wait_time'),
  workersQueueLatency: meter.createObservableGauge('content_workers_queue_latency')
} as const

// later on
const percentiles = ['p50', 'p99', 'p99_9'] as const
metrics.workersQueueWaitTime.addCallback((result) => {
  for (const percentile of percentiles) {
    result.observe(this.workers.waitTime[percentile], { percentile })
  }
})
metrics.workersQueueLatency.addCallback((result) => {
  for (const percentile of percentiles) {
    result.observe(this.workers.runTime[percentile], { percentile })
  }
})

However over the lifetime of the pool, those histogram are never reset so the p99 and p99_9 take a lot of time to get back to a current value since it stores old one. I would like to be able to call this.workers.runTime.reset() to clear the buckets after each value collection.

If that's something you are okay with, i can make a PR for it

[metcoder95]

Just out of curiosity, what would be the shape of the API?
Just to reset the histograms at a given point in time?

Along with that, do you have a distributed setup? How do you measure the distribution among hosts?

[github-actions]

This issue has been marked as stale because it has been opened 30 days without activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue was closed because it has been stalled for 5 days with no activity.

[github-actions]

This issue has been marked as stale because it has been opened 30 days without activity. Remove stale label or comment or this will be closed in 5 days.

[vmarchaud]

@metcoder95 Sorry for some reason i never say your comment

Just out of curiosity, what would be the shape of the API?

i've suggested this.workers.runTime.reset()

Just to reset the histograms at a given point in time?

Yes i don't think it requires something more, it match the underlying implementation used (https://nodejs.org/api/perf_hooks.html#histogramreset)

Along with that, do you have a distributed setup? How do you measure the distribution among hosts?

Yes, prometheus is actually pretty good at re-constructing histogram across different instances